Revisiting obligatory relatives in German

Abstract In the literature on relative clauses (e. g. Alexiadou et al. 2000: 4), it is occasionally observed that the German complex definite determiner d-jenige (roughly ‘the one’) must share company with a restrictive relative clause, in contrast to bare determiners der/die/das (Roehrs 2006: 213–215; Gunkel 2006; Gunkel 2007). Previous works such as Sternefeld (2008: 378–379) and Blümel (2011) treat the relative clause as a complement of D to account for its mandatory occurrence. While such syntactic analyses have intuitive appeal, they pose problems for a compositional semantic analysis. The goal of this paper is twofold. First, we report on two rating studies providing empirical evidence for the obligatoriness of relative clauses in German DPs introduced by the complex determiner d-jenige. Secondly, following Simonenko (2014, 2015), we provide an analysis of the phenomenon at the syntax-semantics interface that captures familiar (Blümel 2011) as well as novel related observations. Particularly, the analysis accounts for the facts that postnominal modifiers can figure in d-jenige-DPs and that the element can have anaphoric demonstrative pronominal uses.


Introduction
The literature on relative clauses in German occasionally observes that there is a special definite determiner or demonstrative-like element dasjenige (roughly 'the neut one', 1 henceforth DJ) which must co-occur with a restrictive relative clause (rRC), (1-a). 2 This contrasts with DPs introduced by other, simplex determiners e. g. der 'the' / ein 'a ' / jeder 'every' in (1-b), where an rRC is optional: ( gross. tall 'The/A/Every man (who stood at the river) was tall.' The empirical problem DJ poses is easily stated: How can we capture the cooccurrence pattern of DJ and rRC, given that rRCs are optional when co-occurring with all other determiners and thus usually treated as adjuncts? The main purpose of this paper is to aim at a comprehensive description of the facts surrounding German DJ. We believe that any analysis of DJ-rRCs must capture the facts to follow, some of which have not hitherto been observed, let alone been analyzed. Any analysis should in addition be part of a fuller theory of DPs which captures optional rRCs like in (1-b) as well. An even better result would be one in which a compositional semantics is part of the treatment -and we provide such an analysis.
Despite the appearance that the phenomenon represents a small empirical corner which one might generously neglect, DJ-DPs must be understood as part of a rich and intricate cross-linguistic pattern and within a larger theoretical endeavor going back to observations of at least Smith (1964). E. g. Gunkel (2006) lays out systematic co-occurrence patterns of determiners and rRCs in Dutch, German, Swedish and French. As for its theoretical relevance and scope, researchers strive to come to terms with e. g. the syntax and semantics of demonstratives and plain definite DPs (cf. Leu 2008), and the ability of the former to allow for bound readings by quantifiers (cf. Simonenko 2015). Furthermore, the current study informs our understanding of relativization in particular, and nominal modification more generally (e. g. Cinque 2015 for one recent approach). So next to contributing to a descriptively adequate account of the DJ-facts, various independent phenomena converge on the formal analysis that we will pursue. This article is structured as follows. In Section 2 we present two rating experiments that validate the observational claims in the theoretical literature regarding DJ-DPs. An additional finding is an asymmetry with respect to the acceptability of extraposition of rRCs: with DJ-DPs rRC-placement in the Nachfeld is judged equally good as their non-extraposed counterparts, while extraposed rRCs were judged significantly worse than their non-extraposed variants with DPs headed by simplex determiners. Devoted to theoretical issues, Section 3 rehashes previous purely syntactic takes on DJ-DPs and proposes the current analysis. It makes use of an account at the syntax-semantics interface by Simonenko (2014Simonenko ( , 2015, developed on the basis of independent facts and considerations, capturing core observations on DJ by Blümel (2011). Extending the findings of the two previous sections, Section 4 first shows that the theoretical analysis from Section 3 allows us to make predictions for novel data. Secondly, it addresses the experimental finding pertaining to extraposition and proposes a tentative explanation in terms of parsing. We hope that this will inspire further research on the intricate and ill-understood facts (Büring and Hartmann 1997). Section 5 concludes.

Experimental perspective
The literature on obligatory rRCs in the context of DJ relies mainly on linguists' individual intuitions -an indispensable source of data and tool for linguistics. Still, in the following, we will provide additional empirical evidence for this, based on two rating experiments on the complex determiner DJ in comparison to their bare determiner counterparts der/die/das.

Experiment 1
Design and stimuli Experiment 1 used a 2*2 factorial design with the factors "Det" (i. e. bare vs. com-plex=DJ) and 'relative clause' henceforth, rRC (+/−, i. e. present or absent). The four conditions are listed in Table 1. We had two predictions: First, sentences containing the complex determiner would be rated more natural with relative clauses than without. Second, there would be no difference in the case of bare determiners. We used a total of 24 items in the experiment. An example of the stimuli is given in (2). The complete list of the test items are provided in Appendix A. In addition, we used 84 fillers consisting of both natural and unnatural sentences.

Procedure and subjects
The stimuli were split up into four counterbalanced sets, each with 108 sentences. The experiment was conducted online at https://www.soscisurvey.de. 36 German native speakers participated in the experiment for credit points. Subjects were instructed to rate the sentences as either natural or unnatural. The real experiment followed a practice stage of 8 sentences, which was intended to familiarize them with the rating task.

Data analysis and result
All analyses were performed using mixed effects linear regression models. The models were constructed using the lme4 package in R (Bates et al. 2015). The reported models are the maximal models that converged. The first model included Det (dem/DJ) and rRC (+rRC/-rRC) as fixed effects. Furthermore, we included random intercepts for subjects and items, as well as random by-subject slopes for the effects of Det and rRC (with interaction). We found a significant interaction between Det and rRC (t = 8.92, p < 0.0001) and a significant effect of Det (t = 5.56, p < 0.0001). As shown in Figure 1, DJ was rated significantly more natural with rRC than without (t = 6.67, p < 0.0001). For simplex definite determiners like dem, on the other hand, sentences with complement clauses (i. e. -rRC) were rated more natural than with rRCs (t = 5.76, p < 0.0001). Based on this, we can confirm the observation that +jenige-sentences without an rRC are degraded. The result that bare determiners with rRCs were rated less natural than without was unexpected. We came up with two possible explanations: First, the DP e. g. der Vorschlag 'the suggestion' and the like might prefer the presence of a complement clause. In line with this was the fact that Condition 3 (bare Det -rRC) satisfying this condition was rated the best among all the conditions. The fact that Condition 2 (complex Det + rRC) with potentially the same problem was rated better than Condition 1 (bare Det + rRC) (t = 3.33, p < 0.05) can be explained as follows: In Condition 2, subjects were happy that the grammatical (syntactic and semantic) requirements of DJ were satisfied, thus paying less attention to the DP, e. g. der Vorschlag, that would occur more naturally with a CP complement (see Fabricius-Hansen and von Stechow 1989). The second possibility is that readers may have dispreferred the distant rRC as it could not be anticipated with the bare determiner. This is compatible with the fact that rRC conditions were rated significantly more natural with the complex determiner DJ, as there the rRC is highly expected. To test the plausibility of these two possibilities, we conducted a followup experiment as described below.

Experiment 2
Design and stimuli Experiment 2 used a 2*2 factorial design with the factors "Det" (bare vs. complex) and 'distance of the relative clause' (distant or local), which, crossed with each other, yielded four conditions, as in Table 2. We used a total of 24 items in the experiment. Crucially, we added a postnominal genitive attribute to the object DP, e. g. der Vorschlag, to satisfy the possible pragmatic constraint that it prefers a complement. An example of the stimuli is given in (3). The complete list of the test items are provided in Appendix B. In addition, we used 96 fillers consisting of both natural and unnatural sentences.
( Our predictions were: First, based on the finding in Experiment 1, there would be no differences between the local vs. distant rRC condition for the complex determiner. Second, in the case of the bare determiner, either there would be no difference between the local vs. distant rRC condition if the degradedness of Condition 1 in Experiment 1 was indeed due to the absence of a complement CP for the DP as e. g. der Vorschlag. Or, the distant rRC condition would still be rated as less natural than the local rRC condition. Speaking in favor of this, Strunk (2014), for instance, reports on experimental findings that DPs without DJ disprefers extraposition, whereas extraposition is fine for DJ.

Procedure and subjects
40 German native speakers were asked to give naturalness judgments on the stimuli, which were split up into four counterbalanced sets. Each subject saw 120 sentences in total. The procedure of the experiment is the same as that of Experiment 1. The real experiment followed a practice stage of 8 sentences, which was intended to familiarize them with the rating task.

Data analysis and results
All analyses were performed using mixed effects linear regression models. The model was constructed using the lme4 package in R. The reported model is the maximal models that converged. The model included Det (dem/DJ) and rRC-Type (rRC local/rRC distant) as fixed effects. Furthermore, we included random intercepts for subjects and items, as well as random by-subject and by-item slopes for the effects of Det and rRC (with interaction). We found a significant interaction between Det and rRC (t = 4.92, p < 0.0001): as Figure 2 shows, while for DJ, both local and distant rRCs were rated equally natural, for dem, distant rRCs were rated significantly worse than local rRCs (t = 6.93, p < 0.0001). Local rRCs with the bare determiner did not differ significantly from either local or distant rRCs with the complex determiner, i. e. all three conditions are equally natural. Distant rRCs, on the other hand, are significantly worse with a bare, than with a complex determiner (t = 5.75, p < 0.0001).

Summary
To sum up, the two experiments provide empirical evidence that the complex determiner e. g. DJ requires a relative clause, and thus confirms the related observation made in the theoretical literature. In addition, they also show that extraposition is optional with the complex determiner whereas the bare determiner seems to have a locality preference. In the following, we will focus on the rRC requirement of the complex determiner and propose a formal analysis for this in Section 3, and return to the related issue of extraposition in Section 4.

Formal analysis
This section provides an analysis of DJ-DPs. We will begin by briefly reviewing previous analytical approaches to DJ-DPs (Alexiadou et al. 2000: 4;Sternefeld 2008: 378-379;Blümel 2011). We then lay out our theoretical assumptions and their empirical motivation. We will adopt the analysis by Simonenko (2014Simonenko ( , 2015 for DJ-DPs and show that it not only helps to understand their various prop-erties previously observed in the literature, but also sheds light on unreported facts.

Previous analyses
To the best of our knowledge, all accounts aiming at coming to terms with the obligatoriness of the rRC in DJ-DPs (repeated here as (4)) capture the phenomenon in selectional, i. e. in syntactic, terms: D directly Merges with the relative clause CP. Assuming that a given head can maximally select one complement, this means that the head-NP cannot be a complement of D. (4) Derjenige DJ  (Sternefeld 2008: 378-379) In the much-debated raising approach to rRCs shown in (5-a), the determiner and the head-noun of the rRC do not form a constituent. The head-noun undergoes raising within the CP-internal, Ā-moved DP, swapping its position with the relative pronoun. (5-b) represents more of a sketch of an analysis rather than a worked-out proposal. The relative-CP resembles an attributive adjective in that its underlying base configuration is prenominal, while the head-NP occupies the right specifier position within the DP. Thus, a proviso must be made to obtain the surface linear order, such as a rightward DP-internal extraposition rule. (5-b) has some intuitive appeal insofar as certain morphological agreement facts of the German DP (ϕ-and Case-agreement) are captured or at least need no special treatment, unlike in the raising account.
There are numerous problems with both these approaches and its descendants (cf. for example, Blümel 2011) but we hold the view that a fundamental problem lies in their reliance on syntactic structure alone. That is, while we believe that the idiosyncratic behavior of the construction is related to syntactic properties of the noun phrase, this cannot be the whole story. The relationship between the special determiner and the obligatory relative clause is more indirect, as we will suggest below.
Syntactic and morphological problems of the head raising analysis in (5-a) abound, and we will here not revisit familiar points of critique that have been raised (cf. Borsley 1997;Heck 2005and, recently, Webelhuth et al. 2018). Likewise, we will not address attempts at repairing some of these defects (Bianchi 1999;De Vries 2002), since we believe these to involve ad hoc solutions. Regarding (5-b), suffice it to say that this proposal raises questions for a compositional mapping between syntax and semantics: in what is arguably the standard treatment of rRCs at the syntax-semantics interface (Heim and Kratzer 1998), the semantic composition of the meaning of a head noun and an rRC comes about by Predicate Modification, which is defined over the denotation of two constituents of type <e, t>. In (5-b), Predicate Modification cannot apply, because the rRC-CP and the NP are not sisters. Assuming that the semantic type of D=dasjenige is <<e, t>, e> (cf. Heim and Kratzer 1998: 75 on English the), applying Function Application to the determiner and the rRC eventually yields a type clash: feeding D' e to the denotation of the specifier, NP <e,t> , delivers the denotation of type t for the DP -the wrong result. Needless to say that the syntactic treatment (5-b) is ad hoc and insufficiently general, designed for only one particular rRC-construction. It arguably entails a non-uniform treatment of relativized DPs in German if one were to integrate optional rRCs and their structure into the general picture.
That said, the empirical problem remains of capturing obligatory rRCs in DJ-DPs. In the following, we would like to suggest a solution which essentially resorts to analytical means which have been proposed and established independently. While the internal structure of rRCs is not the focus of this paper, we would like to briefly point out that the proposed analysis of the DP-structure and its compositional semantics are compatible with a matching approach to relatives (Chomsky 1965;Sauerland 1998), and possibly with a head external analysis (Chomsky 1977;Jackendoff 1977). A raising analysis encounters e. g. the problem that the head noun must vacate the RC in a "sideward" fashion, i. e. end up in a position from which it fails to c-command its trace and violate the Extension Condition (Chomsky 1995). Thus, to the extent that the current analysis, which aims at capturing the relationship between determiners and RCs, is on the right track, it has repercussions for the internal structure of RCs in that it speaks against a raising analysis (and is compatible with alternative approaches in the field).

Background: A DP-internal relational predicate
Austro-Bavarian makes a formal distinction between so-called weak and strong definite articles, exemplified in (6-a) and (6-b) respectively:  (6), the superlative adjective guarantees that the NP denote a singleton, i. e. the one individual that satisfies being simultaneously a mountain and of the type such that no other individual is higher. Taking these observations as a starting point, Simonenko (2015) treats (6-b) as a reflex of an anti-uniqueness presupposition which is part of the semantics of the strong definite article. Whenever the NP-denotation is a singleton, this presupposition is not met and the truth conditions of the sentence the DP is part of cannot be calculated. English demonstratives exhibit a similar restriction: a. the/#that highest mountain b. #That sun is bright. (Simonenko 2015: 196) Simonenko (2015) observes that importantly, the anti-uniqueness presupposition arising in the context of strong definite articles is obviated in the presence of rRCs, here exemplified with Austro-Bavarian again:  (Simonenko 2015: 197) Both NP-denotations involve a singleton in that there can be at most one day on which the speaker is born. Crucially, the strong definite article is compatible with a rRC (8-b), surprisingly contrasting with (6-b). This contrast suggests that the rRC does not play the same role as (superlative) adjectives. Simonenko (2015: 197) thus concludes that "the head carrying the anti-uniqueness presupposition has to have access to the NP to the exclusion of the restrictive relative." In her treatment, the relationship between the definite determiners (strong definite articles, demonstratives, DJ) and the said anti-uniqueness presupposition is rather indirect in that the head introducing the presupposition differs from the D-head which hosts the definite determiner. For strong definite articles and demonstratives she proposes a DP-structure in which a functional projection she calls relational phrase (RP) is hierarchically sandwiched between D and NP, 3 shown in (9): By assumption, R is phonologically null. Crucially, the semantics of R introduces the anti-uniqueness presupposition, and an rRC occupies the specifier position of this relational head. Adjectives adjoin to NP, which means that adjectives are in the syntactic scope of R. This, then, is the reason (6-b) and (7) (with a demonstrative) are bad. (6-a), by contrast, is well-formed, because weak definite articles involve a DP-structure without RP and thus no R-head with an anti-uniqueness presupposition. Given these considerations, we can turn to (8). In the absence of RP, rRCs adjoin to NP in (8-a), as is standard. By contrast, rRCs occupy SPEC-RP in DPs like (8-b). As the rRC is not within the scope of R, R 's anti-uniqueness presupposition simply does not apply to the denotation of the rRC (but only to the NP denotation). This then derives the puzzling asymmetry between attributive adjectives and rRCs with respect to the anti-uniqueness presupposition associated with strong definite determiners and demonstratives. Regarding its compositional semantic contribution, R encodes lexically, what the extrinsic rule Predicate Modification does in the treatment of rRCs by Heim and Kratzer (1998): R 's denotation in (10) is a function applying to "two properties of individuals and, if defined, returns a set of individuals with both properties." (Simonenko 2015: 198) The anti-uniqueness presupposition is a domain restriction on the first argument of R , i. e. P, demanding that the cardinality of P be greater than 1: Since the anti-uniqueness presupposition applies only to the first argument of R , material in SPEC-RP is exempt. Consequently, the denotation (10) allows for the formation of the intersection of (non-singleton) P and (unrestricted) Q. The application of (10) to both arguments delivers the denotation of RP, a set of individuals. The rest of the semantic composition proceeds as is standard, i. e. with a standard semantics for D involving a ι-operator, as Simonenko shows: Notice that R 's anti-uniqueness presupposition applies to the NP-denotation only, and not to the DP-denotation introduced by definite articles (strong and weak alike), i. e. the standard uniqueness presupposition of the definite D-head is retained: While the denotation of R' has to involve a set whose cardinality is greater than one, that set intersects with the denotation of the rRC. The result of that function application meets the definite D-heads' presupposition only if it yields a singleton. DJ can spell-out D iff D takes RP as a complement.

Adopting the proposal
4 Alternative implementations are conceivable, like e. g. a head movement or lowering approach from R to D or D to R respectively. Since the demonstrative d-as-jen-ig-e (d-agr-distal-adj-agr) exhibits a transparent morphological complexity such an approach might have some plausibility. However, in present day German no element can intervene between the individual elements of DJ and so we retain a "conservative" lexical approach.
The obligatoriness of rRC follows from the fact that R applies to two arguments, each of which is a set of individuals. Slightly altering the rendering of the structure suggested by Simonenko, we suggest a rightward specifier 5 so as to capture the fact that the rRC linearly follows rather than precedes the head noun. In the current discussion, nothing hinges on this choice, and alternatives involving Simononko's original proposal plus DP-internal rightward extraposition are readily conceivable. Notice that the analysis adopted in (12-b) makes the claim that the strong/weak distinction observable in Austro-Bavarian dialects has a reflex in standard German: While the definite determiner die 'the' / 'that' is lexically ambiguous between a definite article and a demonstrative, DJ is unambiguously the standard German strong definite determiner resembling the counterpart in Austro-Bavarian. We take it that the transparently demonstrative nature of DJ, which comprises the distal component jen 'that,' 6 supports this treatment and justifies a differential syntactic treatment of simplex determiners and complex, demonstrative(-like) elements like DJ.
In this context, notice that phonological contraction of prepositions and definite determiners is known to be confined to 'weak' definites (14-a) and impossible with 'strong' definites (14-b). Schwarz (2009) claims that the latter are anaphoric. (14) a. reveal 'At every train station that our train entered a letter from the person was handed to me who turned out to be the mayor.' Such facts buttress the basic idea that nominals involving DJ involve a more complex DP-structure than the one of simplex articles and that DJ is demonstrative in nature. We will later adduce more evidence in favor of this conclusion. For now, let us establish that in contrast to the structure underlying DJ-DPs, the analysis of optional rRCs like (16-a) is the usual one, i. e. they adjoin at the NP-level (16-b): Crucially, (13) rules out the realization of D as DJ in such structures as (16-b).
Taking stock, the fact that rRCs in German are usually optional receives the standard analysis: they adjoin to NP in the presence of all other determiners, safe DJ. Moreover, the standard semantics of relatives can be retained for DJ under the present approach, as shown above for strong definite articles. The seemingly exceptional behavior of rRCs in the presence of DJ is captured as well.
Remember that the treatment by Simonenko (2014Simonenko ( , 2015 is a lexical one in that semantically, the rRC is the second argument required by the R -head. Such an analysis raises questions with respect to a well-known property of rRCs: How do we capture the fact that they can be stacked, i. e. they can be iterated? Syntactic operations like adjunction and semantic ones like Predicate Modification, both of which are recursive operations, appear to be indispensable. Consider (17) By exhausting the combinatoric potential of the functor R , a lexical analysis limits the number of relative clauses within the RP to one, leading to an undergeneration problem. However, we would like to argue that (17) provides an argument for (rather than against) the current analysis. Predicate Modification and adjunction remain available and computationally viable operations. Both of these apply at the NP-level, as is standard, while the rRC in SPEC-RP terminates the addition of rRCs. The semantic analysis predicts that in (17) the anti-uniqueness presupposition applies to the first rRC (call it P) and not the second (call it Q). This is indeed correct: P must comprise at least two individuals, otherwise the second relative clause is interpreted non-restrictively. 7 Q forms an intersection with (anti-unique) P. Crucially, P must not be a subset of Q, i. e. there must be individuals in P which are not members of Q. Put differently, P must retain and comprise individuals which function as alternatives to the set singled out by Q. We believe that these considerations deserve more elaboration and thought than we can provide here. In particular, the issue should be addressed whether stacking relatives always involves a syntactic-semantic analysis as the one that we have sketched here, irrespective of the presence of DJ.
Summing up, we have adopted the analysis by Simonenko (2014Simonenko ( , 2015 to accommodate familiar facts of obligatory rRCs in the context of DJ (Alexiadou et al. 2000;Sternefeld 2008;Blümel 2011). Syntactically, the analysis involves a functional projection between NP and DP, i. e. relational predicate layer RP. Semantically, R is a function that takes two sets of individuals and, if the presupposition of the first argument is defined, returns a set of individuals with both these properties. The second argument of R in SPEC-RP is the obligatory rRC. That is, the rRC's occurrence is the result of a lexical requirement; optionally adding rRCs recursively at the NP-level beyond this requirement is still possible. As we will show in the next section, the analysis has numerous desirable empirical consequences.
In the following, we will highlight a few ramifications of the analysis in Section 3. In Section 4.1, we will first demonstrate that the account not only nicely captures DJ-DPs with rRCs, but also DJ-DPs with other modifiers as reported in Blümel (2011), as has also been shown in Simonenko (2014). In Section 4.2, we will provide a further extension of the analysis to account for a set of hitherto unobserved and unexplained facts regarding DJ in pronominal use without any modifiers. In Section 4.3, we will make a tentative proposal regarding Dutch light-headed relatives with the counterpart of German DJ. In Section 4.4, we conclude with a short remark on asymmetries in the extraposition behavior of rRCs with and without DJ, picking up results from the experiments from Section 2. Blümel (2011: 27) observes that postnominal PP-and genitive DP-modifiers can appear in DJ-DPs instead of the rRC, as shown in (18) and (19)  'My sister's.' (Blümel 2011: 27) As before with rRCs, omission of the modifiers yields ungrammaticality (with a qualification to which we return):  (Blümel 2011: 27) These facts suggest an obvious extension of the analysis: the SPEC-RP-position, normally occupied by the rRC, can alternatively host PP-or genitive DP-modifiers.

DJ with other modifiers
That is, constructions involving these modifiers are structurally ambiguous between a standard structure where the modifier is NP-internal and the analysis under current investigation in which these modifiers fulfill the role of the rRC in SPEC-RP: Notice that this analysis requires postnominal (partitive) genitive DPs to denote properties of individuals, something that might be needed on independent grounds. This analysis makes immediate sense of the following fact noted by Blümel (2011: 27): it is insufficient to merely add a modifier of the semantic type <e, t> to DJ to fulfill the determiner's semantic needs. If that were the case, we would expect adding an attributive adjective to DJ to be good, contrary to fact:  (Blümel 2011: 27) Under the analysis in (22) this amounts to saying that N-ellipsis is obligatory. This would be unusual. The counterparts with a simplex article are fine (25), indicating that it is not the redundancy of repeating the NP-part which is the culprit in (24) We cannot solve this puzzle at this point but would like to point the following: The current approach represents a step forward in that it accommodates more data surrounding DJ-DPs, particularly obligatory modifier types other than rRCs.

R in pronominals
There emerge further new predictions of the analysis. Strong definite determiners can figure in readings in which they are bound by quantified DPs if modified by rRCs, as Simonenko (2015) shows. Next to this usage, she suggests that an analysis is available in which instead of an rRC, a silent individual pronoun can occupy SPEC-RP as shown in (26), termed i. This analysis aims at capturing deictic uses of the strong definite determiner in Austro-Bavarian; the referent of i is specified by a contextually given individual: R is of the semantic type <et, <et, et>>. When the DP is used in a deictic fashion, i. e. when SPEC-R is occupied by a silent individual pronoun instead of the rRC, a type clash with the second argument of R must be avoided as illustrated in the tree above. To do so, Simonenko (2015) adopts the type shifting operation ident (Partee 1987), turning the individual pronoun into a set of individuals: We adopt that analysis to the DJ-case. SPEC-RP can alternatively host either an individual pronoun, an rRC or a PP/DP-modifier, which functions as the second argument of R . This predicts the following: We expect that DJ can function as a demonstrative pronoun, anaphoric to a discourse antecedent. 9 This is borne out, as a corpus search reveals. 10 In the examples (28), the discourse antecedent is boldfaced: 9 The fact that demonstrative pronominal DJ prefers discourse-anaphoric uses and resists deictic uses has to remain an issue for future research. Assuming that non-rRCs adjoin to DP only, the current analysis accounts for this in that a missing SPEC-R leads to a failure to satisfy the second semantic argument of R , resulting in a type clash higher up in the structure. 11 The account also gives rise to another novel prediction: non-rRC can co-occur with DJ after all, iff DJ has a discourse antecedent. The reason is that in this context, the structure is that of a deictic demonstrative in which R must take the silent individual pronoun i as a second argument. This makes SPEC-RP unavailable for an rRC.
There is then nothing that prevents the introduction of a non-rRC, which adjoins to the DP. Once more, the facts fall into place: decided 'This means that no one must get his/her organs extracted unless that person, who must have been examined, has given permission during his/her lifetime.' The grammaticality of (30) is thus on a par with a DJ-DP co-occurring with both an rRC and a non-rRC (which is acceptable as well, as expected).

Dutch light-headed relatives
Dutch features a counterpart of German DJ, het-and datgene respectively. Unlike in German, omitting the head is mandatory in Dutch (cf. Gunkel 2006: 4- (Blümel 2011: 28) Such "light-headed relatives" are quite common cross-linguistically and are attested in e. g. Romance (French, Italian, Spanish) as well as Polish (cf. Citko 2004).
We would like to suggest that Dutch light-headed relatives involve a simpler structure than the one considered for German. Specifically, we would like to claim that the structure of Dutch light-headed relatives is a simple complement of D-analysis, immediately capturing the impossibility of a head-noun.
There is thus no need for a uniform analysis of German and Dutch DJ-facts: None is warranted by the data.

A short remark on extraposition
In Section 2, we reported on two rating experiments on DJ-DP in comparison to DPs headed by bare determiners. They confirm the observation made in the theoretical literature that DJ-DPs need a relative clause, which we formally analyzed in Section 3. In addition, we also found an asymmetry in the extraposition behavior of rRCs: these were judged as significantly worse than local rRCs for the bare determiners, while no such difference obtained for DJ-DPs. While relative clause extraposition was not the main focus of our paper, we will discuss it briefly in this section.
First, Strunk (2014) reports on corpus studies on relative clause extraposition showing that it can be influenced by multiple factors such as the distance of extraposition, the length of the relative clause, and the specific type of determiner used. He conducted a rating experiment on the bare vs. complex determiner (i. e. DJ) in German. Based on a 2*2 design, he used stimuli as in (36). One of his findings was that distant rRCs were judged significantly worse than local rRCs in the case of bare determiners, whereas there was no difference between distant vs. local rRCs in the case of DJ. 'Jens laughed at the/that musician after the concert who couldn't play the easy piece.' While the design (see Section 2) and the stimuli of our experiments, see (3), differ from Strunk's (2014), the results are similar to his. Taken together, this means that the bare determiner has a locality preference which lacks with DJ. The question is why. While we do not have a final answer to this, any answer should take into consideration the morphological makeup of the DJ, as in (37)  The stem of the distal demonstrative is crucial here. It makes the DJ an anaphoric demonstrative, which requires a relative clause to provide a "property" antecedent. A discourse context with an antecedent as in (28) can serve the same function, but relative clauses are the most common strategy to satisfy this requirement. This explains why local or distant rRCs were judged as equally good for the DJ. In contrast, a bare determiner such as der is ambiguous between a plain definite article as its common use vs. an anaphoric demonstrative with an antecedent in discourse context (cf. Schwarz 2009). In other words, bare determiners by their semantics do not require a forthcoming relative clause. In terms of language production, speakers would put parts of a sentence (i. e. an rRC and the hosting NP) that belong together semantically next to each other, all things being equal (see also the related discussion in Strunk 2014: 89). In terms of comprehension, the human parser integrates a local rRC immediately into the DP, resolving the referent. A distant rRC, however, requires reanalysis of the already resolved referent and thus can be perceived as less natural. This view is in line with the Discourse Locality Theory by Gibson (1998) which assumes that integrating a new discourse event (in this case, by an rRC) incurs structural integration costs, which depend on the distance from its dependents. Taken these altogether, this suggests that the lower naturalness rating of "bare Det + distant rRC" might reflect extra integration costs. DJ is different in this regard due to its underlying syntax and semantics, such as the one that we have proposed here, which triggers the expectation of an rRC, no matter if local or distant.
A final remark is that there might be cross-linguistic variation concerning DPs that require rRCs with respect to locality. For example, the English DP the one requires a relative clause, as shown in (38-a). According to our native speaker informants, (38-a) is a well-formed sentence but not (38-b). If this observation is valid in general, this means that the one has a locality preference for the rRC. Whether it is the case also with common DPs e. g. the student is an empirical question. How does the processing of relative clauses for these differ from that of German DJ-DPs vs. bare DPs? We will leave this for future comparative studies.

(38)
These past few days, I interviewed many students.
a. I hired the one *(who struck me as most competent) as a tutor. b. ??I hired the one as a tutor who struck me as most competent.
In this paper, we have reconsidered the observation in the theoretical literature that the complex determiner d-jenige (roughly 'the one') in German requires a restrictive relative clause. We reported on two rating experiments showing the complex determiner without a relative clause results in degradedness and thus providing empirical evidence for the observation. A secondary finding from the two experiments is that the relative clause is equally natural being local to or extraposed from the complex determiner. In contrast, bare determiners prefer a local relative clause over a distant one.
Focussing on the requirement of d-jenige for an obligatory relative clause, we discussed problems of existing syntactic accounts for the phenomenon and proposed a compositional analysis by treating the relative as a semantic argument of a relational head R, following Simonenko (2014Simonenko ( , 2015. We argued that this analysis captures the German data from the literature (Alexiadou et al. 2000;Sternefeld 2008;Blümel 2011) and extends naturally to new observations in German and, possibly, beyond.